• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

50TB+ flash array/storage you recently purchased

M

mp3turbo

n00b
Joined
Dec 25, 2013
Messages
16
Guys,

pretty soon, we might start to look for "price CONSCIOUS" flash storage with approx. 60TB usable capacity right now with expansion to 90TB expected in a year. Will be used for data warehousing appz on 20 ESX v5.5 hosts, 10Gbit FCoE, single location, single room. Apologies for not mentioning detailed environment specification as I don't consider it to be necessary now [see below, please, before shooting me and don't talk about redundancy].

It does NOT have to have the best-of-the-world-all-bells-and-whistles, ie. we can survive without dedup. Basically I can't recall any really sub-par flash vendor as of features, so I'm not worried too much. Usually if a vendor X is missing this or that functionality, they will add it within 6 to 12 months which is extremely fine with us on general basis. Of course, careful consideration / comparison will be performed.

Before I run into every flash vendor in the world, I would like to hear your experience as of configuration/price. I'd be interested in info like "we procured ABC with xx TB raw capacity, performance specified as xxxx IOPS / yyyy MB/s for $$$$$$ and all that happened in October 2014". That will easily show me where to look and this is the pure reason why I'm not mentioning utter details of our environment.

Big thanks in advance. Feel free to post additional questions, but you generally see what I'm after. The list prices are useless and we can't deal straight away with 15+ suppliers as it's time consuming and we simply don't have the resources.

A note : we are not interested in any SSD caching, tiers etc in standard arrays as their architecture generally is not as capable as all-flash arrays have. Also we believe there is no relevant price & performance advantage for us, we checked with vendors of storages we already use - knocked my socks off when I've heard the price - so we'd rather buy entirely new flash storage for this DWH, effectively freeing up existing resources. This was cross-posted in SSD/Storage forum as I don't know which is better.
 
End to end FCoE or FCoE to the host, FC to the array? I'm not aware of any all flash products that offer FCoE natively. Pure, XtremIO, 3Par, and Tegile don't. I'm guessing you have a Nexus 5k or something with hosts using CNAs so you can use an array with iSCSI or FC.

Honestly, your costs are going to be all over the board. Not only will vendor X be more expensive than vendor Y at the get go, but once vendor X hears they're more money they'll come down in price and then so will vendor Y. I've seen a deal where vendor X was $1.4M and vendor Y was $400k for roughly the same capacity and performance. Vendor X came down from $1.4M to match the $400k!

Pure tends to be best bang for the buck. Solidfire is also a sound product and worth looking at. XtremIO is awesome now that v3 is shipping.

What is "price conscious?" What are you looking to spend?

Also, out of curiosity, how have you determined that all 60-90TB require the performance of flash?
 
Cheers for prompt reply and three vendor names (what a fiasco with XtremIO destructive update, on the other hand they have been very closely cooperating with affected customers) ! Prices being all over the place is exactly the reason I created this topic.

FCoE : we are already separating it to 10Gbit Ethernet (iSCSI) and 8Gbit FC so the new storage easily can use "standard interfaces".

As of spending, we are into six figures with "the less the better" as usually. Customer does OLTP, OLAP, DWH, DSS, and is getting into situation where spinning disks are really no longer able to cope with those loads - they will get to maximum disk configuration of current storage pretty soon with IOPS being the limit.

It has already been decided they want 35TB usable for the most critical and performance sensitive data right from the start. Sure there are other very easy targets for bombastic performance. I'm just afraid 60TB I said is too little :) everybody hopes the effect of oflloading current storage will be worth it too.

I always had the feeling that EMC is extremely expensive with their XtremIO, given the roots they came from and the clientele they historically had. Hmmmm. Heard EMC is particularly concerned about Pure, to the extent their sales people have anti-Pure talking points. That makes me curious, sure. Last time I checked XtremIO - about mid-2014, it was limited to 120TB raw in maximum brick configuration which is no-go for us [moreover given the price]. Yes I hate the compression/data reduction unrealistic miracles vendors claim for their products.
 
Unless you plan on creating lots of Dev/Test/QA clones of the DBs I wouldn't count on more than a 2:1 dedup/compression ratio for databases.

I would talk to Pure about an array. I've sold them in the past and every customer has been ecstatic with them. They tend to be very competitive price wise and are incredibly easy to set up and use. For 60TB usable you'd be looking at the FA-420 model. Can't give you a price estimate since Pure sells through the channel but you could get a ballpark figure by estimating $5/GB or less.

As for XtremIO, if you can get a competitive situation going where one partner is offering Pure and the other XtremIO you really can't go wrong no matter which solution you choose. Both are phenomenal products.
 
Would you mind if I reached out?

I can get you what you need, at the same performance as Pure, for about 1/2 the price, and all in a single 4U chassis.

Oh, and I'll make it so you can spin up an entire test/dev copy of any given DB/app system in less than 2 minutes, with zero overhead...
 
lopoetve said:
Would you mind if I reached out?

I can get you what you need, at the same performance as Pure, for about 1/2 the price, and all in a single 4U chassis.

Oh, and I'll make it so you can spin up an entire test/dev copy of any given DB/app system in less than 2 minutes, with zero overhead...
Click to expand...

Share with the crowd who/what you do / work with. :cool:
 
So I used to work for VMware (should be pretty obvious from my prior posts ;)) - spent almost 7 years there as an architect on the cloud side, and one of the lead storage engineers for the company.

Now I work for Tintri. What we provide is a hybrid array, but it's not like anything else on the market (I guarantee it - give me an hour of your time on a phone call, and I'll blow your mind). We were doing VVOL-esque management 3 years ago with our first generation product, before VMware even had written the spec for everyone else.

All management is done per VM - replication, clones, snapshots, backups, restores... all done on a VM-by-VM basis, from the vSphere web client (or our client). There's not a single LUN, raid group, or share to manage - it's all 100% per-VM. Just build workloads, instead of configuring arrays and trying to size properly.

Per-VM workload identification and management. We profile each VM by talking to the ESX hosts and Virtual Center to identify what it is, what it's doing, and what the workload is. We literally configure flash and spinning resources to match the workload, regardless of it being SQL, Exchange, Apache, Java, or Big Data.

Per-VM QoS done natively. When at VMware, I cloned and booted over 1500 VMs at once in an attempt to track down and break a Virtual Center scaling issue. The array natively cloned all 1500 VMs (2k8/2k12/RHEL even mix) in 45 seconds, booted them in just over 2.5 minutes, latency hit a max of 2.3ms... and the 100 seat VDI environment that was sharing the array never even noticed. Each VM is wrapped in a set of resource lanes that prevents any one workload from ever negatively impacting other workloads on it. And in a coming update, you'll be able to adjust min/max levels of QoS on the fly, with a simple slider...

Easiest interface to learn. Ever. The setup process has three steps: Rack it, Give it an IP, point it at Virtual Center. Done. Everything else is done on the VMs :)

Full end-to-end latency and performance monitoring, per-VM. We talk to both Virtual Center and to the ESX hosts to get a full, complete picture of end-to-end performance. I can literally track IO from the moment it leaves the guest application till it hits the array, all the way through the entire vSCSI/VMkernel stacks, and the network, plus our entire system. Brocade literally just used it to track down a bug in their new VDX firmware they were seeing - we could tell the moment they hit the right workload size to trigger it, and clearly showed the latency appearing on the network.

Replication, per-VM, with a built in riverbed/data domain type system. Literally right click the VM, set the schedule, and you're done. Ever configured SnapMirror? This is as dead simple as it gets in comparison. If you can deploy a template, you can configure our replication and backup systems.

VMSync. I can literally take a running exchange cluster, clone it (for your average 5-node DAG environment, it takes about 35 seconds), attach it to your dev/test/qa cluster, customize it, and you've now got a dev/test place... and it took 0 space on the array, btw. And, once you're done running tests, if you'd like to copy over the current data to it from the running production environment, I can resynchronize the disks to the current production version, reset space usage to zero, and it takes less than 20 seconds total.

Oh and all those snapshots and clones? First filesystem designed for it from the get go, with no technology borrowed from ZFS/WAFL/CASL/etc. Sub-microsecond performance hit for each. Or from garbage collection (we built in a GC portion of the filesystem). Or from RAID rebuilds (now that's a fun story!). Oh, and feel free to fill it up. Unlike many of those, we don't really take a performance hit from getting full - you'll lose about 5% at 95% full, but that's it. Try that with some of our competitors, and it's more like 50-60% at 85-90% full... ;)

All these features? Also available on Hyper-V, or KVM, or soon, OpenStack. And we're contributing our Cinder driver back to the openstack community once it's done. And I'm even going to give you the ability to translate, natively on the array, between VMware, Hyper-V, and (probably, still working on this) KVM.

Top model, I can get you 180k non-normalized IOPS, up to 100T usable, sub-1ms latency up to that 180k limit, and all in a 4U chassis with 1/2 the cost of Pure or xtremeIO.

I know, it sounds like a sales spiel and too good to be true - but I'm a tech person. I went here from VMware because I believed in what they were doing, and everything I ~tried~ to do, it simply did. And I never had to waste time managing the array. When I left, our 540 (3 years old now) was running several production cloud environments, a production exchange cluster, 3 VDI pools, and a bunch of test/dev work... and each group had no idea they were sharing an array, since no one could impact anyone else ;)

If you want to know more, let me know. Give me an hour of your time to walk through our technology and show you a demo, and if you're not impressed, I'll buy you lunch somehow. I'm not joking.
 
i'm personally evaluating both pure and xtremeio as I type this. I'm not buying the compression and dedupe hype in either of them. XtremeIO has some serious short falls, IE.. have to take it offline to add space to it, or upgrade the firmware. Pure doesn't suffer from that. But Pure is still fairly new to the game. So is xtrememio, but they are also backed by EMC. However, EMC is shaking because pure is eating into some serious market share. People are moving off of V-Max over to pure for raw performance.

The main reason why I haven't personally been pushing all flash arrays in my environment is because there are still to many different vendors all trying to do the same thing. I'm waiting on the herd to be thinned out.. 2nd, very few if any all flash arrays support 2TB drives today this second. Its on the road map, but until its actually got a skew and shipping, I'm holding off. 2TB efds are the tipping point for me, in making it worth wild in getting off 2TB+ sata drives.

@OP.. All Flash Array, and Price conscious do not go into the same sentence. With out knowing a whole lot about your environment, actual iop, and response time demands, its pretty tough to answer your question.. BUT.. I just implemented an EMC VNX 5400 with around 150+ ts of space, tiering, and flash cache that took up 1/3rd of a rack for a killer deal that actually replaced a fully popped NS480 that took up 3.5 racks of space. VNX can do all EFD/Flash drives so its not a true flash array. Still hybrid. But it gives me quite a bit of flexibly in the future as 4tb sata drives are still cheap. edit.. and it was cheaper then an all flash array and im getting 60K iops out of it.
 
Last edited:
kdh said:
XtremeIO has some serious short falls, IE.. have to take it offline to add space to it, or upgrade the firmware.
Click to expand...

Sort of. You can expand a starter half brick to a full 10TB brick without downtime and you can upgrade the firmware without downtime (not 2.4 -> 3.0 however, that is disruptive). Adding an X-Brick to an existing cluster is disruptive.

Shortfalls, definitely, but not as bad as a simple firmware upgrade requiring downtime. The 3.0 upgrade only required downtime because they had to change the array block size to add inline compression.

I, too, tend to steer towards Pure because it doesn't suffer from these problems and tends to cost less.
 
I actually thing both products are killer. It'll be a shoot out for sure. Honestly, I wouldn't be shocked if EMC didn't just buy pure, and take what they like out of it, and add it to xtremeio.

I just racked a 5tb brick of xtreme io on Friday. Haven't powered it on yet, but the first thing I'll do is get it to the 3.0 code base and go from there. I may FTS it behind my v-max and play with it that way as well.
 
lopoetve said:
Now I work for Tintri.
Click to expand...

OMG you left the mothership!?

Also, ignore PM I just sent, your post explains it.

@OP
XtremeIO fiasco aside, EMC is more than aggressive with pricing on XtremeIO for new flash array customers. The whole "vendor X does it for less" doesn't really apply when you look at the raw per/TB pricing. Depending on your existing relationship with EMC the capex for XtremeIO is equal if not lower than the capex of competitors if you compare raw TB.

It all comes down to features and/or less-than-tangible things. For example, in a 100% Cisco and EMC environment, do you want to add a third vendor to your stack, even if there are some things that the third vendor could potentially do better? At the 60TB you are asking for the answer is likely a resounding NO. Unification matters. How much that matters to you only you and your people can decide.
 
Thuleman said:
It all comes down to features and/or less-than-tangible things. For example, in a 100% Cisco and EMC environment, do you want to add a third vendor to your stack, even if there are some things that the third vendor could potentially do better? At the 60TB you are asking for the answer is likely a resounding NO. Unification matters. How much that matters to you only you and your people can decide.
Click to expand...

Im struggling with this exact issue as I type this. I handle roughly 98% of the storage here.. Your talking 3 V-Max installs, vnx 2 5400s, 2 5300s, 2 5700s, a chumpy ns-120, 3 Isilon clusters, 3 Data Domain 860s fully popped, all running on a cisco san. In fact, I only wanted Clariion/VNX and V-Max in my environment. I had to flex and do Isilon because Isilon is everything Celera ever wanted to be and did what I needed it to do. Now with all these flash vendors on the map? Surprisingly, Hitachi does not have a Flash array. They have one on the road map but its build from VSP parts.

So the idea of bringing another hardware stack(dedicated flash array) into my environment? Regardless or not if its EMC or another Vendor.. I'm not happy about it.
 
kdh said:
Im struggling with this exact issue as I type this. I handle roughly 98% of the storage here.. Your talking 3 V-Max installs, vnx 2 5400s, 2 5300s, 2 5700s, a chumpy ns-120, 3 Isilon clusters, 3 Data Domain 860s fully popped, all running on a cisco san. In fact, I only wanted Clariion/VNX and V-Max in my environment. I had to flex and do Isilon because Isilon is everything Celera ever wanted to be and did what I needed it to do. Now with all these flash vendors on the map? Surprisingly, Hitachi does not have a Flash array. They have one on the road map but its build from VSP parts.

So the idea of bringing another hardware stack(dedicated flash array) into my environment? Regardless or not if its EMC or another Vendor.. I'm not happy about it.
Click to expand...

What if you never had to manage it?

Seriously - I had a T540 for 2 years. I logged into it 3 times in those two years, and once was during the initial setup. All the rest? Just building and managing VMs.

Don't believe me? I'll see about getting you an eval unit, if you'll give me an hour of your time - if you have to manage it 1/4 of the time of any of your other devices, I'll buy you lunch.
 
Thuleman said:
OMG you left the mothership!?

Also, ignore PM I just sent, your post explains it.

@OP
XtremeIO fiasco aside, EMC is more than aggressive with pricing on XtremeIO for new flash array customers. The whole "vendor X does it for less" doesn't really apply when you look at the raw per/TB pricing. Depending on your existing relationship with EMC the capex for XtremeIO is equal if not lower than the capex of competitors if you compare raw TB.

It all comes down to features and/or less-than-tangible things. For example, in a 100% Cisco and EMC environment, do you want to add a third vendor to your stack, even if there are some things that the third vendor could potentially do better? At the 60TB you are asking for the answer is likely a resounding NO. Unification matters. How much that matters to you only you and your people can decide.
Click to expand...

I'll make you the same offer. Give me an hour of your time on the phone, and if you're not impressed... I'll buy lunch.
 
(Disclaimer: I work for HP but I’m not paid to sell anything!)

I think for any hybrid or all-flash situation, HP 3PAR should be in your evaluation. HP 3PAR was just named the AFA product of the year by TechTarget. My colleague and HP blogger Calvin Zito just posted an article talking about this award and what sets HP 3PAR apart.

Here are a few things he mentioned:
• The performance of the 3PAR AFA is an industry leader, including start-up AFAs. A general guideline we've talked about is 900,000 IOPs at under .7 ms latency. Our read latency is as low as 300 micro-seconds and write latency as low as 200 micro-seconds. Yes, that was MICRO-seconds.

• Deduplication with 3PAR is done via our Gen4 ASIC. That means dedup happens at virtually "line speed". Most other AFAs dedup using the same CPUs that do other tasks and during heavy IO workloads, they turn off deduplication. You said that you aren’t interested in deduplication – this is an important feature to include as it will lower the cost of your solution.

• Another cost savings comes from the 3PAR architecture; we have a unique advantage with AFA. SSD vendors reserve space on drives to aid in ware management. HP 3PAR has a unique feature called Adaptive Sparing that allows 3PAR to increase the size of SSDs by 20% - so a 1.6TB SSD used by other AFAs is a 1.9TB drive with HP 3PAR.

• Our converged flash array, 3PAR 7440c, can drive several hundreds of thousands of IOPs at under .2ms. This is notable as it shows that with 3PAR using HDDs and SSD, we're seeing better performance than the industry average for AFAs today.

There is a really good (with technical details) article on the blog titled Built for Flash that helps explain why 3PAR is so ideally suited for flash.

I’m happy to get you in touch with an expert that can answer any questions.

Isaac.
 
I'll settle this. Both sales reps gimme free shit to run at the house :D
 
lopoetve said:
So I used to work for VMware (should be pretty obvious from my prior posts ;)) - spent almost 7 years there as an architect on the cloud side, and one of the lead storage engineers for the company.

Now I work for Tintri. What we provide is a hybrid array, but it's not like anything else on the market (I guarantee it - give me an hour of your time on a phone call, and I'll blow your mind). We were doing VVOL-esque management 3 years ago with our first generation product, before VMware even had written the spec for everyone else.

All management is done per VM - replication, clones, snapshots, backups, restores... all done on a VM-by-VM basis, from the vSphere web client (or our client). There's not a single LUN, raid group, or share to manage - it's all 100% per-VM. Just build workloads, instead of configuring arrays and trying to size properly.

Per-VM workload identification and management. We profile each VM by talking to the ESX hosts and Virtual Center to identify what it is, what it's doing, and what the workload is. We literally configure flash and spinning resources to match the workload, regardless of it being SQL, Exchange, Apache, Java, or Big Data.

Per-VM QoS done natively. When at VMware, I cloned and booted over 1500 VMs at once in an attempt to track down and break a Virtual Center scaling issue. The array natively cloned all 1500 VMs (2k8/2k12/RHEL even mix) in 45 seconds, booted them in just over 2.5 minutes, latency hit a max of 2.3ms... and the 100 seat VDI environment that was sharing the array never even noticed. Each VM is wrapped in a set of resource lanes that prevents any one workload from ever negatively impacting other workloads on it. And in a coming update, you'll be able to adjust min/max levels of QoS on the fly, with a simple slider...

Easiest interface to learn. Ever. The setup process has three steps: Rack it, Give it an IP, point it at Virtual Center. Done. Everything else is done on the VMs :)

Full end-to-end latency and performance monitoring, per-VM. We talk to both Virtual Center and to the ESX hosts to get a full, complete picture of end-to-end performance. I can literally track IO from the moment it leaves the guest application till it hits the array, all the way through the entire vSCSI/VMkernel stacks, and the network, plus our entire system. Brocade literally just used it to track down a bug in their new VDX firmware they were seeing - we could tell the moment they hit the right workload size to trigger it, and clearly showed the latency appearing on the network.

Replication, per-VM, with a built in riverbed/data domain type system. Literally right click the VM, set the schedule, and you're done. Ever configured SnapMirror? This is as dead simple as it gets in comparison. If you can deploy a template, you can configure our replication and backup systems.

VMSync. I can literally take a running exchange cluster, clone it (for your average 5-node DAG environment, it takes about 35 seconds), attach it to your dev/test/qa cluster, customize it, and you've now got a dev/test place... and it took 0 space on the array, btw. And, once you're done running tests, if you'd like to copy over the current data to it from the running production environment, I can resynchronize the disks to the current production version, reset space usage to zero, and it takes less than 20 seconds total.

Oh and all those snapshots and clones? First filesystem designed for it from the get go, with no technology borrowed from ZFS/WAFL/CASL/etc. Sub-microsecond performance hit for each. Or from garbage collection (we built in a GC portion of the filesystem). Or from RAID rebuilds (now that's a fun story!). Oh, and feel free to fill it up. Unlike many of those, we don't really take a performance hit from getting full - you'll lose about 5% at 95% full, but that's it. Try that with some of our competitors, and it's more like 50-60% at 85-90% full... ;)

All these features? Also available on Hyper-V, or KVM, or soon, OpenStack. And we're contributing our Cinder driver back to the openstack community once it's done. And I'm even going to give you the ability to translate, natively on the array, between VMware, Hyper-V, and (probably, still working on this) KVM.

Top model, I can get you 180k non-normalized IOPS, up to 100T usable, sub-1ms latency up to that 180k limit, and all in a 4U chassis with 1/2 the cost of Pure or xtremeIO.

I know, it sounds like a sales spiel and too good to be true - but I'm a tech person. I went here from VMware because I believed in what they were doing, and everything I ~tried~ to do, it simply did. And I never had to waste time managing the array. When I left, our 540 (3 years old now) was running several production cloud environments, a production exchange cluster, 3 VDI pools, and a bunch of test/dev work... and each group had no idea they were sharing an array, since no one could impact anyone else ;)

If you want to know more, let me know. Give me an hour of your time to walk through our technology and show you a demo, and if you're not impressed, I'll buy you lunch somehow. I'm not joking.
Click to expand...


OMG you left the mothership!?

I can literally take a running exchange cluster, clone it (for your average 5-node DAG environment, it takes about 35 seconds), attach it to your dev/test/qa cluster, customize it, and you've now got a dev/test place... and it took 0 space on the array, btw. And, once you're done running tests, if you'd like to copy over the current data to it from the running production environment, I can resynchronize the disks to the current production version, reset space usage to zero, and it takes less than 20 seconds total.
Click to expand...


Interesting Stuff. I am still reading....this is good information.
 
lopoetve said:
I'll make you the same offer. Give me an hour of your time on the phone, and if you're not impressed... I'll buy lunch.
Click to expand...

What I am saying though is that it's not always about the technology but about everything else that goes along with it.

In our environment Tintri could perform twice as good as XtremeIO and cost half of it, and we would still not buy it. It's not about whether it's impressive, it's about how it diversifies our environment, how it would lower the discounts we receive from EMC, how there is an additional vendor to call if there are issues and vendors have the option to blame each other, etc. etc.

The one thing I learned by switching jobs myself recently is that capex and even dollar amounts of TCO are no longer the actually relevant numbers when it comes to decision making.

This isn't even a Tintri vs other vendor observation, it's a generic one vendor vs another vendor observation.
 
lopoetve said:
What if you never had to manage it?

Seriously - I had a T540 for 2 years. I logged into it 3 times in those two years, and once was during the initial setup. All the rest? Just building and managing VMs.

Don't believe me? I'll see about getting you an eval unit, if you'll give me an hour of your time - if you have to manage it 1/4 of the time of any of your other devices, I'll buy you lunch.
Click to expand...

I hear what you are saying. The 5tb brick of xtremeio that I have, is a play ground that I can mess with. I completely believe that both xtrememio and pure, are both very simple and easy to setup and maintain. I get that part, and totally believe it. The part that takes the most time is actually migrating the data to it. My use case isn't for VMware, but for a very highly randomize oracle OLTP DB in the financial industry. The DB is over 30Ts in size. I want to explain more about what I do, but I just don't want to in a public forum. But moving from one array to another while sounds trivial, there is always "gotchas" that you have to be ready for. 98% for that migration is ASM rebalance which is a no brainer, but I do a lot of boot from san with UCS, and logical volumes. I have to san copy my boot luns, and I would use powerpath migrator for the logical volume stuff. Both of those 2 process's can be a little hairy and be the most time consuming. In my environment, if I don't have the time to properly eval a new flash array and I have to stand something up quickly, I will either just pile more space into my V-Max arrays or VNX arrays because I know them to operate exactly how I expect them to operate. Again, financial industry and I need to 100% standby what I say we can do when asked. This makes sense today. But in the long term? I will have to radically change the way I do things. Obviously evaluating a few platforms is key, and then figuring out what makes the most sense in my environment.

Spinning disk is long in the tooth.. It aint got much more time left.

But trust me.. I love the fact that I can just cut up my luns, provision them to the host, and then I'm done. No performance tuning or tweaking, no fast or fast vp policies to work with. The only thing I have to worry about is FA port saturation. I would still have to learn its management tools, and performance monitoring tools. But hey, they can't be any worse then unisphere for vmax. ;)
 
NetJunkie said:
Getting a little heavy on the sales leads here guys... ;)
Click to expand...

I KNOW, RIGHT?

3Par my have some amazing stats, but like every single vendor out there, they all have hero type scenarios, with unicorn type work loads. Obviously to get those numbers, you have the most perfect data scenario absolutely possible.

But guess who else does that same exact thing? EMC, Hitachi, Netapp.. Hell, lets really make it nutty, Oracle Exadata is doing the same thing and is really beating on my door to replace my UCS and V-Max with next gen Exadata.
 
Thuleman said:
What I am saying though is that it's not always about the technology but about everything else that goes along with it.
Click to expand...

This is so on point.

Sales guys love to say how much time will be saved when switching to a different stack.. But really how much time is REALLY saved? Racking and Stacking, powering it up, and cutting some luns is a no brainer.. its CAKE. More so with all flash arrays. Its alllllllll the other stuff that goes with it. Upgrading it, Adding space to it, trouble shooting bottle necks, understanding the management tools, how to monitor its performance, Being able to tell if its a host performance issue or an array issue and migration data over to it. Hell.. even boot from san stuff. How that works on V-Max, and VNX could be radically different with a different vendor.

I also know exactly how my arrays will function, with a new stack, there is a bit of a gray area for a few months until you get comfortable with it. Sometimes having that known grantee is worth it for me to stay on current stack, then chance it and goto a different stack. Again.. in my situation, financial industry.. I have to make absolutely sure that a solution I implement doesn't knock over my environment. Because no one likes it when you play with their money.

2596126-9801756839-draft.jpg
 
Last edited:
sc0tty8 said:
I'll settle this. Both sales reps gimme free shit to run at the house :D
Click to expand...

I love this, and totally spit my soda out when I read this. =)

Hope power is cheap where you live cause your electric bill is gonna go up 10xs. So at that point, is it REALLY free? Lets take this off the rails.. Lets start talking OPEX vs CAPEX for your house. *head explodes*
 
kdh said:
I hear what you are saying. The 5tb brick of xtremeio that I have, is a play ground that I can mess with. I completely believe that both xtrememio and pure, are both very simple and easy to setup and maintain. I get that part, and totally believe it. The part that takes the most time is actually migrating the data to it. My use case isn't for VMware, but for a very highly randomize oracle OLTP DB in the financial industry. The DB is over 30Ts in size. I want to explain more about what I do, but I just don't want to in a public forum. But moving from one array to another while sounds trivial, there is always "gotchas" that you have to be ready for. 98% for that migration is ASM rebalance which is a no brainer, but I do a lot of boot from san with UCS, and logical volumes. I have to san copy my boot luns, and I would use powerpath migrator for the logical volume stuff. Both of those 2 process's can be a little hairy and be the most time consuming. In my environment, if I don't have the time to properly eval a new flash array and I have to stand something up quickly, I will either just pile more space into my V-Max arrays or VNX arrays because I know them to operate exactly how I expect them to operate. Again, financial industry and I need to 100% standby what I say we can do when asked. This makes sense today. But in the long term? I will have to radically change the way I do things. Obviously evaluating a few platforms is key, and then figuring out what makes the most sense in my environment.

Spinning disk is long in the tooth.. It aint got much more time left.

But trust me.. I love the fact that I can just cut up my luns, provision them to the host, and then I'm done. No performance tuning or tweaking, no fast or fast vp policies to work with. The only thing I have to worry about is FA port saturation. I would still have to learn its management tools, and performance monitoring tools. But hey, they can't be any worse then unisphere for vmax. ;)
Click to expand...

Well, we couldn't help you anyway - you can't hook physical boxes up to a Tintri, it's virtualization-only. :)
 
You must log in or register to reply here.
Back
Top